Yash Vyas
Cell: +1-510-***-**** Email: ac0jo8@r.postjobfree.com Linkedin: yashvyas
Education: Stanford University, MS in Statistics (expected graduation: June 2017); GPA 3.52/4.0 IIT Madras- India, B.Tech. in Civil Engineering and M.Tech. in Applied Mechanics, 2012 Technical Skills: R, Python, TensorFlow
Interests: Statistics, Data Mining, Machine Learning, Deep Learning, NLP Research and Internships
Optimization of cellular networks using machine learning models (Summer Internship at Uhana, Palo Alto) (June-Sept, 2016)
Developed time-series models to make recommendations with confidence intervals for optimizing cell user experience in real time
Clustered mobile towers in a city based on traffic and consumption pattern using dynamic time warping
Recovered prominent features from unstructured data and developed a prediction model with 20% mean percent error rate
Developed a model that penalizes asymmetrically on errors made in pre-positioning content and maintain false accepts and false rejects rates at < 5% and < 20% respectively
Auto-suggesting values to fields in documentation (Research Assistant at Stanford Medicine School) (March-Dec, 2016)
Developed a 400-class classification model from text-descriptors to predict outputs having long-tail distribution with 75% accuracy
Improved prediction accuracy by dynamically updating recommended values based on user inputs during the documentation
Generated features from unstructured repositories and identified similar clusters in the output using topic-models Sensitivity analysis of cause of tuberculosis spread (Centre for Applicable Mathematics-TIFR, Bangalore) (Sept 2014-Feb 2015)
Conducted uncertainty and sensitivity analysis for a tuberculosis prevalence model using Latin Hypercube Sampling to identify the effects correlations among the inputs on the model output
Estimated the effect of parameter variability on the disease load and assisted the Indian government in the disease-survey Projects
Machine Comprehension Using Deep Learning (Class project) (Jan-March, 2017)
Developed a bi-directional attention flow model using LSTM in TensorFlow to predict answer sequence within an input context using context-query text tuple in the SQuAD dataset
Predicted answers with exact match accuracy of 39% and F1 score of 0.50 on unseen context - query data tuples Fake Review Detection on Yelp Restaurants (Class project) (Sept-Dec, 2016)
Developed a Bayesian approach to detect fraud based on features extracted from reviews, author account and product listings
Obtained an F1 score of 0.56 and overall accuracy of 91% on classifying fake reviews using a dataset having ~12% fake reviews
Used generalized methods such as Tf-Idf, Latent Dirichlet allocation and word2vec to capture behavioral traits in fake reviews Prediction of progress of ALS disease (Class project) (Sept 2015-Dec 2015)
Modeled the relationship between rate of progress of disease and biological features using regression and classification techniques
Improved upon the existing model to predict the disease progress and finished 7th out of 165 teams in the Kaggle competition Chances of Winning in Indian Parliamentary elections (Independent project) (Jan-May 2014)
Predicted winning probabilities for parliamentary candidates using a regression model including candidate popularity, party history at the constituency, available opinion polls and existing legislative assembly results in the 2014 General Elections, India
Projected the voting trend and correctly identified the winning candidates before the election results for 162 out of 201 seats Relevant Professional Experience
Quantitative Analyst, Labs Team, Tookitaki- Bangalore, India (Feb 2015-July 2015)
Developed temporal similarity measures among text entities using current news and trending topics on the web
Used words similar to actual keyword in ads for reaching the comparable audience groups at 50% cheaper rates
Designed a method to efficiently allocate budget among different digital marketing campaigns and improved conversion rates Publication
Stochastic Creep Damage Growth due to Random Thermal Fluctuations using Continuum Damage Mechanics, Creep, Fatigue and Creep-Fatigue Interaction, Procedia Engineering, Elsevier (2013)